Machine Learning Framework for Antalgic Gait Recognition Based on Human Activity

Antalgic gait is one of the most common abnormalities in human beings during the walking. This work presents a framework for the automatic recognition for antalgic and non-antalgic gaits, using the gyroscope of a smartphone for data acquisition. The test carried out was 10-meter walk, with a population of 30 subjects, 40% antalgics, and 60% non-antalgics; 80% was used in the training stage, and the rest for the test. A hypothesis testing and p-value method were developed to determine the statistical difference between both datasets and validate the usefulness of data in the features selection and classification approach. The classification algorithms used were: i) K-Nearest Neighbors (k-NN), ii) Naive Bayes (NB), iii) Support Vector Machines (SVM), iv) Discriminant Analysis (DA), v) Decision Trees (DT), and vi) Classification Ensembles (CE). The performance of the algorithms was evaluated using the metrics: Accuracy (ACC), Sensitivity (R), Specificity (SP), Precision (P), and F-measure (F). k-NN and SVM were the models with better performance with Accuracy of 99.44% and 98.88%, respectively. The obtained results allow to determine the feasibility of implementing this framework in real scenarios for its use in the improvement of diseases diagnosis and decision-making to antalgic gait diseases.